1 Motivation and Goals

Predicting 30-day readmission on the basis of surgery/procedure characteristics
Visualizing disease-disease networks
Successfully identifying those diagnoses and procedures that can be targeted for quality improvement

2 What we have

Medicare data representing inpatient claims at the national level.
About 350,000 records of inpatient visits paid by Medicare
5 to 10 diagnoses per inpatient record
Demographic data such as age, race, income quartile of patient zip code
You can learn more about Medicare data here
American College of Surgeons’ National Surgical Quality Improvement Program (NSQIP)
High-fidelity data, ACS NSQIP identifies more 30-day outcomes than do other quality programs
Considers 30-day complications, which form the basis for Affordable Care Act reimbursement

Size of node represents frequency of the diagnosis among all diagnoses
Thickness of the edge connecting two nodes represents ‘lift’, which is how much more often we see the pair than we would expect if they occurred independently
Black represents procedure-specific probability of 1 that the patient will have an unplanned readmission to the hospital, whereas increasingly brighter shades represent lower probability of unplanned readmission.

Yellow indicates more days spent in the hospital for that procedure, while purple indicates fewer days spent in hospital.

Confusion Matrix and Statistics

Metrics
Accuracy	0.6311
95% CI	(0.6263, 0.6358)

Confusion Matrix and Statistics

Metrics
Accuracy	0.6287
95% CI	(0.6191, 0.6383)

Size of node represents frequency of the diagnosis among all diagnoses
Thickness of the edge connecting two nodes represents ‘lift’, which is how much more often we see the pair than we would expect if they occurred independently
Selected relationships based on co-incidence metrics
Black represents about 1 day spent at the hospital, whereas increasingly brighter shades represent greater length of stay.

Thanks!